My job is to build systems that work. Code is a liability sometimes needed to accomplish that. I don't want more of it unless there's some reason to believe the benefits outweigh the liability. Based on what I'm able to get from LLMs and especially what I'm seeing produced by others I know are using them, that's not often the case with agents.

> Why isn’t that a win?

Because, as others have put it, "amount of written code" is more akin to "weight of an airplane" than "productivity"

That of course isn’t what I meant. I make sure the code I produce isn’t inefficient or verbose.

Because that’s precisely what’s wrong with AI. It creates a lot of code, what could have been done with fewer code, or no code at all if you didn’t go on path A with AI, but thought about path B yourself.

That strictly depends on your ability to direct it precisely with accurate prompts.